Score: 34.7 | Matched keywords: agent, ai, ai agent, llm, prompt, reasoning, token
Categories: cs.SE, cs.AI
When an AI agent calls an API and hits a validation error, it needs more than what went wrong -- it needs what to do next. A self-reflective API returns, on validation failure, a machine-readable recovery_feedback.suggestions[] payload sufficient for the agent to repair the request and retry without external reasoning.
When an AI agent calls an API and hits a validation error, it needs more than what went wrong -- it needs what to do next.
A self-reflective API returns, on validation failure, a machine-readable recovery_feedback.suggestions[] payload sufficient for the agent to repair the request and retry without external reasoning. On a leak-audited pilot (N{=}30 per cell, 3 LLMs, 10 adversarial tasks), structured suggestions lift task-completion rate by +36.7--40.0 pp over plain-English diagnoses on Anthropic models (Fisher's exact p 0.0022), at 1.8--2.2 better per-success token efficiency.
The lift is not significant on gpt-4 o-mini (p{=}0.435); a second-domain replication on a billing API confirms the pattern. The comparison only holds after auditing two undocumented classes of answer leakage in LLM benchmarks.
When an AI agent calls an API and hits a validation error, it needs more than what went wrong -- it needs what to do next. A self-reflective API returns, on validation failure, a machine-readable recovery_feedback.suggestions[] payload sufficient for the agent to repair the request and retry without external reasoning. On a leak-audited pilot (N{=}30 per cell, 3 LLMs, 10 adversarial tasks), structured suggestions lift task-completion rate by +36.7--40.0 pp over plain-English diagnoses on Anthropic models (Fisher's exact p 0.0022), at 1.8--2.2 better per-success token efficiency. The lift is not significant on gpt-4 o-mini (p{=}0.435); a second-domain replication on a billing API confirms the pattern. The comparison only holds after auditing two undocumented classes of answer leakage in LLM benchmarks. We shipaudit_prompt_leakage.py as reusable CI infrastructure. Code and data: https://github.com/arquicanedo/self-reflective-apis.